The Causal Relationships Between Inflammatory Proteins, Brain Structure, and Psychiatric Disorders: A Two-Step Mendelian Randomization Analysis

Abstract

Background and Hypothesis

Inflammatory proteins are implicated in psychiatric disorders, but the causality and underlying mechanisms remain unclear.

Study Design

We conducted bidirectional Mendelian randomization (MR) using genetic variants from genome-wide association studies (GWAS) for 91 inflammatory proteins (N = 14 824) and 11 psychiatric disorders (N = 9725 to 1 035 760). The primary analysis used the inverse variance weighted (IVW) method, with additional sensitivity analyses to confirm robustness. A two-step MR approach assessed whether brain imaging-derived phenotypes (IDPs) mediated the observed effects.

Study Results

Forward MR analysis found the protective effect of CD40 on schizophrenia (SCZ) (IVW OR = 0.90, P = 5.29 × 10⁻⁶) and bipolar disorder (BD) (IVW OR = 0.89, P = 5.08 × 10⁻⁶). Reverse MR demonstrated that increased genetic risk of Tourette’s syndrome (TS) was associated with reduced Fms-associated tyrosine kinase 3 ligand (Flt3L) levels (Flt3L) (Wald Ratio beta = −0.42, P = 1.99 × 10⁻⁷). The protective effect of CD40 on SCZ was partially mediated by the modulation of fractional anisotropy (FA) values in the right and left superior frontal occipital fasciculus, with mediation proportions of 9.6% (P = .025) and 11.5% (P = .023), respectively.

Conclusion

CD40 exerts an immunoprotective effect on SCZ and BD, and the effect of CD40 on SCZ was partially mediated through modulation of FA values in the superior frontal occipital fasciculus. These findings enhance comprehension of the etiology of these psychiatric conditions and underscore the promise of therapeutic strategies aimed at inflammatory proteins.

Introduction

Psychiatric disorders encompass a spectrum of debilitating mental illnesses characterized by disruptions in brain function, leading to disturbances in cognitive, emotional, and behavioral functioning in affected individuals. Nearly one billion people worldwide suffer from psychiatric disorders.¹ Due to high rates of incidence, disability, and mortality,² psychiatric disorders have been identified as one of the top ten major burdens worldwide¹ and are considered a major public health concern on a global scale. Elucidating the mechanisms underlying psychiatric disorders is one of the great challenges in psychology and neuroscience. Inflammatory processes play a pivotal role in maintaining normal physiological activities within the body, and aberrant inflammatory responses can lead to impaired brain development and function, which has received attention as a potential pathogenesis.

There is growing evidence for inflammation involvement in the pathophysiology of psychiatric disorders. Infections during pregnancy^3–5 and early childhood,⁶ in addition to autoimmune diseases^7,8 have been linked to an elevated risk of various psychiatric disorders, such as mood disorders and schizophrenia-like behavioral disorders. Numerous observational studies have reported altered levels of inflammatory proteins, including inflammatory cytokines, immune-regulatory proteins, and acute-phase reactants in blood^9–12 and cerebrospinal fluid^13,14 of patients with psychosis. In addition, antipsychotic drugs¹⁵ and antidepressants¹⁶ can modulate peripheral inflammatory protein levels, with these alterations being linked to treatment outcomes.¹⁷

Despite extensive evidence linking inflammation and psychiatric disorders, the causal relationship remains scarce. Furthermore, current conclusions predominantly come from traditional observational studies, which are susceptible to confounding factors and reverse causality. Therefore, it is crucial to explore the causal relationship between inflammatory proteins and psychiatric disorders using a reliable method. Mendelian randomization (MR) serves as a novel approach for examining causal inferences through the utilization of genetic variants as instrumental variables.¹⁸ These variants are randomized during meiosis, providing a form of natural randomization that reflects the principles of randomized controlled trials.¹⁹

In this study, we evaluated the causal relationships between 91 inflammatory proteins and 11 psychiatric disorders by conducting bidirectional two-sample MR analyses, utilizing data from recent large-scale genome-wide association studies (GWAS). Considering the broad spectrum of psychiatric disorders and the potential role of inflammation across different diagnostic categories, we selected 11 psychiatric disorders covering a range of categories and severities. This selection allowed us to comprehensively explore the relevance and specificity of inflammation in psychiatric disorders, providing a more holistic understanding of the role of inflammation in mental health. Our objective was to determine the direction and magnitude of the impact of inflammatory processes on susceptibility to psychiatric disorders.

Emerging evidence suggests that inflammation may cause structural changes in brain imaging-derived phenotypes (IDPs),²⁰ which could increase the risk of certain psychiatric disorders.²¹ Neuroinflammation has been linked to alterations in brain structure,²² such as changes in gray matter volume²³ and cortical thickness,²⁴ which are known to be associated with various psychiatric disorders.²⁵ Structural brain changes provide a stable and reliable indicator of the long-term effects of inflammation on the brain. And compared to functional brain connectivity, structural changes have a more direct and measurable impact on the connectivity and overall architecture of the brain. Therefore, we proceeded with a two-step analysis to determine if structural brain changes mediate the impact of inflammatory proteins on psychiatric disorders. Our findings may provide insights for preventive strategies and early interventions in psychiatric disorders at the level of immune-inflammation.

Materials and Methods

Study Overview

Figure 1 displays the study workflow. Initially, single-nucleotide polymorphisms (SNPs) derived from the GWAS summary data of 91 inflammatory proteins served as genetic instruments. Subsequently, a bidirectional two-sample MR analyses were performed to evaluate the causal effect of each inflammatory protein on 11 psychiatric disorders, as well as the reverse causality. Finally, a two-step MR analysis was performed to investigate the potential mediating role of brain IDPs in the pathway linking the identified inflammatory protein to the psychiatric disorder.

Figure 1.

Study Workflow. Abbreviations: GWAS, Genome-Wide Association Study; BMI, Body Mass Index; SNP, Single-Nucleotide Polymorphism; IVW, Inverse-Variance Weighted Method; IDPs, Imaging-Derived Phenotypes

Data Sources

Inflammatory Proteins.

The genetic information pertaining to inflammatory proteins was acquired from the recently published GWAS summary statistics of 91 inflammatory proteins.²⁶ This study included 11 cohorts with a combined total of 14 824 participants, all of European ancestry. Plasma protein levels were measured using the Olink Target-96 Inflammation immunoassay panel. The detailed information can be found in Table 1 and Supplementary Table 1.

Table 1.

Characteristics of Included Genome-Wide Association Studies for MR Analysis

Traits	N cases	N controls	N Sample	Ancestry	PubMed ID	GWAS References
Inflammatory proteins	NA	NA	14 824	European	37 563 310	Wainberg M et al.26
Psychiatric disorder
OCD	2688	7037	9725	European	28 761 083	IOCDF-GC and OCGAS27
TS	4819	9488	14 307	European	30 818 990	Yu, D. et al.28
ASD	18 381	27 969	46 350	European	30 804 558	Grove, J. et al.29
ANO	16 992	55 525	72 517	European	31 308 545	Watson, H. J. et al.30
ANX	30 993	69 883	100 876	European	26 754 954	Otowa, T. et al.31
PTSD	23 212	151 447	174 659	European	31 594 949	Nievergelt, C. M. et al.32
ADHD	38 691	186 843	225 534	European	36 702 997	Demontis, D. et al.33
SCZ	55 365	219 884	275 249	European	35 396 580	Trubetskoy, V. et al.34
Insomnia	109 548	277 440	386 988	European	35 835 914	Watanabe, K. et al.35
BD	41 917	371 549	413 466	European	34 002 096	Mullins, N. et al.36
MDD	294 322	741 438	1 035 760	European	37 464 041	Als, T. D. et al.37
Brain structure
Brain IDPs	NA	NA	33 224	European	33 875 891	Smith, S. M. et al.38

Abbreviations: ADHD, attention deficit hyperactivity disorder; ANO, anorexia nervosa; ANX, anxiety disorder; ASD, autism spectrum disorder; BD, bipolar disorder; MDD, major depressive disorder; OCD, obsessive-compulsive disorder; PTSD, post-traumatic stress disorder; SCZ, schizophrenia; TS, tourette’s syndrome.

Psychiatric Disorders.

The genetic information for 11 psychiatric disorders was sourced from publicly accessible GWAS summary data, including obsessive-compulsive disorder (OCD),²⁷ Tourette’s syndrome (TS),²⁸ autism spectrum disorder (ASD),²⁹ anorexia nervosa (ANO),³⁰ anxiety disorders (ANX),³¹ post-traumatic stress disorder (PTSD),³² attention-deficit hyperactivity disorder (ADHD),³³ schizophrenia (SCZ),³⁴ insomnia,³⁵ bipolar disorder (BD)³⁶ and major depressive disorder (MDD).³⁷ These datasets predominantly comprised individuals of European descent, with the number of participants ranging from 9725 to 1 035 760. There was no observed overlap in samples with the study on inflammatory proteins. The detailed information can be found in Table 1.

Brain Imaging-Derived Phenotypes.

Summary statistics for brain IDPs were sourced from Smith et al.,³⁸ utilizing data from 33 224 individuals of European descent in the UK Biobank. The Oxford Brain Imaging Genetics (BIG40) web browser (https://open.win.ox.ac.uk/ukbiobank/big40/) was the source of all GWAS summary statistics related to brain IDPs. Drawing on the robust findings from a prior causal analysis of brain IDPs and psychiatric disorders,²¹ we identified 28 brain IDPs as potential mediators. The detailed information and download links are shown in Table 1 and the Supplementary Table 2.

Genetic Instruments Selection

Selecting Genetic Instruments.

To select appropriate genetic instruments for our study, we applied three rigorous criteria to ensure reliable causal inference.³⁹ The instruments must exhibit a robust association with the exposure under investigation, not independently affect the study outcome, and remain unaffected by confounding variables. First, we selected SNPs that were strongly associated with the exposure (P < 5 × 10⁻⁸). Next, we employed the clump function in PLINK 1.9 to identify independent SNPs linked to each exposure, referencing the 1000 Genomes Project Phase 3 European dataset. We conducted clumping under rigorous conditions, setting the linkage disequilibrium threshold at r² < 0.001 and defining a spatial window of 10 000 kb. We retained SNPs with the lowest P value when they were in linkage disequilibrium. To fulfill the second criteria above, SNPs which showed a strong association with the outcome (P < 5 × 10⁻⁸) were removed. To ensure that beta values aligned to the same allele, we harmonized the effects of SNPs on exposure and outcome. Besides, we removed the palindromic SNPs with nearly 0.5 minor allele frequencies that might cause strand-flipping issues.

Removing Confounders.

To eliminate potential interference, SNPs linked to confounders that disrupt the connection between exposure and outcome should be excluded. Five potential confounders, including education, smoking, body mass index (BMI), autoimmune disease, and cortisol, were considered in the analysis due to previous studies indicating their potential impact on both psychiatric disorders and abnormal inflammation processes.^40,41 We utilized the PhenoScanner V2 database⁴² (http://www.phenoscanner.medschl.cam.ac.uk/) and the NHGRI-EBI GWAS catalog⁴³ (https://www.ebi.ac.uk/gwas/) to remove SNPs that exhibit a significant association with these confounders (P < 5 × 10⁻⁸) in European participants. Finally, 61 confounder SNPs were removed.

Quality Control of Genetic Instruments.

To strengthen the reliability of the remaining genetic instruments, we assessed the robustness of selected genetic instruments by calculating the F-statistics using the formula F = R²  (N − 2)/(1 − R²), where R² = 2 × EAF × (1 − EAF)×β²/[2 × EAF × (1 − EAF)×β²+2 × EAF × (1 − EAF) × SE²×N] (R²: the proportion of exposure explained by instrumental variables, N: GWAS sample size for exposure, EAF: the effect allele frequency, β: the effect size of genetic instrument on exposure, SE: standard error of the effect size).^44,45 The SNPs with F-statistics below 10 were deemed as weak instrumental variables and were therefore removed.⁴⁶ We performed MR analysis on the remaining SNPs and calculated their statistical power based on the method described by Burgess.⁴⁷

Statistical Analyses

Bidirectional Two-Sample MR Analyses.

We conducted bidirectional two-sample MR analyses, employing the inverse variance weighted (IVW) regression method with a multiplicative random-effects model as the primary method for causal inference.¹⁸ Given the potential bias in IVW estimates due to invalid genetic instruments and horizontal pleiotropy, we further conducted additional complementary methods (Weighted median, Weighted mode, Simple mode, and MR-Egger) to assess the robustness of our results. The weighted median method can yield reliable estimates even when as much as 50% of the information is sourced from invalid instrumental variables.⁴⁸ The weighted mode method can produce robust estimates even if the most instrumental variables are invalid.⁴⁹ The MR-Egger method can offer a more dependable estimation even in cases where all variations are invalid.⁵⁰ The Wald ratio was utilized when there was only one instrumental variable available. We expanded the aforementioned MR analysis to investigate bidirectional causality between inflammatory proteins and psychiatric disorders. We applied the Bonferroni correction for all MR analyses to adjust for multiple comparisons, setting the statistical significance threshold at 2.50 × 10⁻⁵ (0.05/91/11/2; 91 denotes the number of inflammatory proteins, 11 denotes the number of psychiatric disorders, and 2 denotes bidirectional MR tests).

Sensitivity Analyses.

To validate the reliability of causal estimates, we carried out multiple sensitivity analyses. The MR-Egger regression⁵⁰ was initially employed to determine the existence of horizontal pleiotropy, with a deviation from zero in the intercept term (P < .05) suggesting a directional pleiotropic bias. Furthermore, the MR-Pleiotropy Residual Sum and Outlier (MR-PRESSO)⁵¹ was utilized to identify pleiotropy (P < .05) and remove outliers. Under the assumption of no horizontal pleiotropy, it evaluates the observed variants by comparing the distances between the variants and the regression line to the expected distances, thereby determining whether these variants exhibit any anomalies. Additionally, Cochran’s Q test⁵² was employed to assess the heterogeneity of individual causal effects within the IVW method. A significance level of P < .05 was considered to indicate the presence of heterogeneity. A leave-one-out analysis was conducted to assess whether any specific variant influenced the causal relationship. Moreover, the application of the MR Steiger test assisted us in validating the assumption that exposure results in the outcome.

Mediation Analysis.

We performed a two-step MR analysis to investigate the potential mediating effect of brain IDPs in the relationship between identified inflammatory proteins and psychiatric disorders. In the first step, we evaluated the causal impact of individual inflammatory proteins on brain IDPs. In the second step, we measured the causal effects of these brain IDPs on the occurrence of psychiatric disorders. We evaluated the indirect effect of a specific inflammatory protein on psychiatric disorder through brain IDPs by using the product of coefficients method.⁵³ The proportion of mediation by specific brain IDPs was ascertained by dividing the indirect effect by the total effect. Standard errors of indirect effect were calculated utilizing the delta method.

Analytical Tools.

The MR analyses were all executed using R software (version 4.3.0), utilizing packages like “TwoSampleMR,” “MendelianRandomization,” “MR-PRESSO,” “RadialMR,” and “RMediation.”

Results

Genetic Instruments Selection

After LD pruning, we used the 1000 Genomes Project Phase 3 European data as reference to select SNPs that were conditionally independent using plink. We removed outlier SNPs (Supplementary Table 3) and SNPs linked to confounders (Supplementary Tables 4 and 5) for subsequent MR analyses. Supplementary Tables 6 and 7 list the full set of SNPs utilized for the forward and reverse MR analyses. The F-statistics were calculated to assess the robustness of genetic instruments. Within the 705 forward and 910 reverse MR pairs that included at least one SNP, all of the F-statistics surpassed the value of 30, indicating the appropriateness of these SNPs as genetic instruments (Supplementary Tables 6 and 7).

Bidirectional Two-Sample MR Analyses

Forward MR Analysis.

The IVW and Wald Ratio analyses identified 52 statistically significant causal relationships at a nominal level (P < .05) (Supplementary Table 8). However, only two causal relationships remained after Bonferroni correction to reach statistical significance (P < 2.50 × 10⁻⁵), as shown in Figure 2. CD40 was negatively associated with the risk of SCZ (IVW odds ratio (OR), 0.90; 95% confidence interval (CI), 0.86 to 0.94; P = 5.29 × 10^-6) and BD (IVW OR = 0.89, 95% CI, 0.85 to 0.94, P = 5.08 × 10⁻⁶). These results were further supported by the weighted median method. Figure 2 displays scatter plots that illustrate the impact of SNPs on the two causal relationships. Various MR analyses are represented by colored lines showing the slopes.

Figure 2.

Causalities in the Forward MR. (A) The Forest Plot Shows the Significant Causalities. The Effect Estimates Represent the Log (OR) of Disorder Per 1 s.d. Change in Mean CD40, and the Error Bars Represent 95% CIs. All Statistical Tests were Two Sided. P < 2.50 × 10⁻⁵ After Bonferroni Correction was Considered Significant. (B) Scatterplot of Single-Nucleotide Polymorphism (SNP) Effects on CD40 vs Schizophrenia (SCZ) and Bipolar Disorder (BD), with the Slope of Each Line Corresponding to Estimated MR Effect Per Method. Data are Expressed as Raw β Values with 95% CI. IVW Indicates the Inverse-Variance Weighted Method

The results of sensitivity analyses performed on the forward MR analysis were shown in Supplementary Table 9. The intercepts of all associations in the MR-Egger regression were near zero, suggesting the absence of pleiotropy. The MR-PRESSO did not identify any indications of horizontal pleiotropy. The results of Cochran’s Q statistics were presented in Supplementary Table 9, indicating that there was no statistically significant evidence of heterogeneity. The leave-one-out analyses demonstrated that the causal estimates were not affected by any individual SNP (Supplementary Figures 1 and 2). Moreover, the MR Steiger test didn’t uncover any signs of reverse causality (Supplementary Table 6).

Reverse MR Analysis.

In reverse MR analysis, no evidence of an association between SCZ (IVW OR, 1.01; 95% CI, 0.97 to 1.04; P = .718) or BD (IVW OR, 0.98; 95% CI, 0.93 to 1.04; P = .575) and CD40 was identified in this study (Supplementary Table 10). The IVW and the Wald Ratio analyses revealed 57 statistically significant causal relationships at a nominal level (P < .05) (Supplementary Table 10). After Bonferroni correction, we found evidence suggesting that the increased genetic risk of TS was linked to a reduction in Fms-associated tyrosine kinase 3 ligand (Flt3L) (Wald Ratio beta = −0.42, 95% CI, −0.58 to −0.26, P = 1.99 × 10⁻⁷) (Supplementary Figure 3).

Furthermore, we conducted a comparison of the IVW results both before and after implementing the confounder filtering to assess for potential unintended bias resulting from the filtering process. The results indicated that the MR estimates are generally similar before and after confounder filtering (Supplementary Table 11). After multiple testing corrections, the significant associations of inflammatory proteins and disorders remained consistent between the two groups.

Mediation Analyses

Considering that some brain IDPs are causally related to SCZ and BD,²¹ and that inflammation may also lead to abnormal changes in some brain IDPs.^20,54 Hence, we further investigated whether the effects of CD40 on SCZ and BD are mediated through modulation of brain IDPs. Twenty-eight brain IDPs were included in this study as potential mediators. We found that elevated levels of CD40 nominally increased the risk of higher fractional anisotropy (FA) values in the right superior fronto-occipital fasciculus (SFOF) (IVW beta, 0.043; 95% CI, 0.001 to 0.084; P = .043) and the left SFOF (IVW beta, 0.041; 95% CI, 0.000 to 0.083; P = .049), although these associations did not pass the Bonferroni correction threshold of 1.8 × 10⁻³ (0.05/28) (Figure 3 and Supplementary Table 12). Conversely, the inverse relationships of FA values in the right and left SFOF with SCZ did survive the Bonferroni correction threshold, with increased FA values in the right SFOF (IVW beta, −0.236; 95% CI, −0.276 to −0.197; P = 1.03 × 10⁻⁵) and FA values in the left SFOF (IVW beta, −0.291; 95% CI, −0.332 to −0.249; P = 1.33 × 10⁻⁵) both negatively causally related to a decreased risk of SCZ (Figure 3). In the mediation analyses, we acknowledge the nominal mediation effect estimates for FA values in the SFOF of both cerebral hemispheres. Specifically, the mediation effect estimate for FA value in the right SFOF was estimated to be −0.010 (95% CI, −0.020 to −0.000), with a mediation proportion of 9.6% (95% CI, 0.3% to 19.2%; P = .025). Similarly, the mediation effect estimate for FA value in the left SFOF was estimated to be −0.012 (95% CI, −0.024 to −0.000), with a mediation proportion of 11.5% (95% CI, 0.06% to 23.1%; P = .023) (Table 2). These mediation effects did not pass the Bonferroni correction threshold but are highlighted for their potential biological relevance.

Figure 3.

Mediation Analysis of CD40 on Schizophrenia (SCZ). (A) Summary MR Estimates Derived from the Inverse-Variance Weighted (IVW), Weighted Median, Weighted Mode, Simple Mode, and MR-Egger Methods for CD40 on “Fractional Anisotropy (FA) in the Right Superior Frontal Occipital Fasciculus (R-SFOF)” (Left) and CD40 on “FA in the Left SFOF (L-SFOF)” (right). (B) Summary MR Estimates for “FA in the R-SFOF” on SCZ (Left) and “FA in the L-SFOF” on SCZ (Right). (C) Visualization Map of Bilateral FA-SFOF

Table 2.

The Mediation Effect of CD40 on SCZ via Affecting Brain IDPs

Mediator	Total effect β (95% CI)	Direct effect A β (95% CI)	Direct effect B β (95% CI)	Mediation effect β (95% CI)	P value	Mediated proportion (%) (95% CI)
FA in the right SFOF	−0.104 (−0.148 to −0.059)	0.043 (0.001 to 0.084)	−0.236 (−0.276 to −0.197)	−0.010 (−0.020 to −0.000)	0.025	9.6 (0.3 to 19.2)
FA in the left SFOF		0.041 (0.000 to 0.083)	−0.291 (−0.332 to −0.249)	−0.012 (−0.024 to −0.000)	0.023	11.5 (0.06 to 23.1)

Abbreviations: IDPs, imaging-derived phenotypes; IVW, inverse variance weighted; SCZ, schizophrenia; SFOF, superior fronto-occipital fasciculus.

Total effect: indicates the effect of CD40 on SCZ; Direct effect A: the effect of CD40 on mediator; Direct effect B: the effect of mediator on SCZ; Mediation effect: the effect of CD40 on SCZ through affecting the mediator. IVW derived the total effect, direct effect A, and direct effect B; the mediation effect was derived by using the delta method. All statistical tests were two-sided. P < .05 was considered significant.

Discussion

This study systematically investigated the causal relationship between 91 inflammatory proteins and 11 psychiatric disorders by utilizing the latest GWAS data. The findings suggest that CD40 exerts a protective effect on SCZ and BD, and TS may lower Flt3L levels through a causal pathway. Moreover, mediation analysis further indicated that FA values in the SFOF partially mediated the effect of CD40 on SCZ. These findings may provide new perspectives for understanding and treating psychiatric disorders.

To our knowledge, this is the most comprehensive MR analyses conducted to investigate the potential association between inflammatory proteins and psychiatric disorders. We found that the protective effect of CD40 on SCZ and BD reached statistical significance. CD40 is a membranous glycoprotein belonging to the tumor necrosis factor receptor superfamily,⁵⁵ and it is broadly present in diverse cell types, including macrophages, lymphocytes and neuronal cells.^55,56 The CD40-CD40 ligand (CD40L) immune-activation pathway is essential for the development of autoimmunity and the host response to infectious pathogens.⁵⁷ Prior studies have demonstrated that the CD40-CD40L pathway is correlated with SCZ and BD. The differentially expressed genes (DEGs) of SCZ patients are mainly enriched in pathways associated with inflammatory mechanisms, with the pathways of highest relevance including the CD40 signaling.⁵⁸ Serum CD40L levels were significantly increased during the development of schizophrenia.⁵⁹ However, in another relevant study, neither SCZ nor BD showed significant changes in CD40L.⁶⁰ One observational research explored the correlation between a range of inflammatory biomarkers and general cognitive ability in these two disorders and found a negative relationship between the levels of CD40L in the serum and general cognitive abilities.⁴⁰ Nevertheless, the constraints inherent in the study design restrict the capacity of these observational correlations to definitively ascertain causality. It is obvious that prior studies have mainly centered on CD40L rather than CD40, largely due to the technical complexities associated with quantifying CD40 levels in circulating blood. Despite CD40L being the ligand for CD40, it is crucial to distinguish their individual impacts. In our investigation, we found a protective association linking CD40 to both SCZ and BD. This observation might be connected to the indispensable roles of the CD40-CD40L interaction in immune reactions and the CD40 signaling pathway in maintaining normal physiological functions of neuronal cells.

The CD40-CD40L interaction has been found to be critical for immune cell processes such as stimulation, maturation, multiplication, and apoptosis regulation.⁶¹ This interaction is vital for the initiation of both cell-mediated and antibody-driven immune responses.^62,63 During the process of host infection, CD40 can enhance the body’s immune response by improving the antigen-presenting capability of dendritic cells.⁶¹ Furthermore, the Type I Interferon (IFN) cytokines are known to activate the body’s inherent and adaptive immunity against viral and bacterial threats.⁶⁴ CD40 emerges as a pivotal facilitator, amplifying the Type I IFN response.⁶⁵ In addition, CD40 is essential for neuronal protection, maintenance, and development in both in vitro and in vivo contexts.⁶⁶ Evidence supports that CD40 ligation shields neuronal cells from damage resulting from serum or nerve growth factor (NGF)-β deprivation and can also regulate neuronal cell development and function by affecting neuronal differentiation.⁶⁶ Animal experiments have revealed that mice lacking the CD40 receptor exhibit impaired neuronal function in adulthood, characterized by a reduction in neurofilament isoforms, alterations in neuronal morphology, increased DNA fragmentation, and overall brain abnormalities.⁶⁶ All these lines of evidence support the protective effect of CD40 against psychiatric disorders.

However, contrary to our conclusion, a recent brain organoid study suggests a potential detrimental impact of elevated CD40 levels in SCZ. Stankovic et al. found that CD40 levels exhibited a general upregulation in SCZ organoids, and in addition, endothelial cells within SCZ brain organoids demonstrate transcriptional alterations related to the regulation of CD40, and the results of a causal network analysis also suggested that CD40 could serve as a shared connection among the genes that exhibit differential regulation in SCZ endothelial cells.⁵⁶ The results of Stankovic et al.’s study also support the idea that neuroinflammation is an inherent manifestation of SCZ during the initial stages of cerebral maturation and that CD40 may ultimately contribute to the pathogenesis of SCZ by affecting endothelial alterations, which in turn impairs blood-brain barrier function. The divergence in results between our study and the brain organoid research may stem from methodological differences and the nature of the samples utilized. While the brain organoid study focused on in vitro SCZ organoid tissues, our MR approach explored the genetic relationships between CD40 and SCZ in a population-based context. Furthermore, the discrepancy could also be attributed to the complex and multifaceted role that CD40 plays in the pathophysiology of SCZ. Given that CD40 is expressed in diverse cell types within the human body, the understanding of its functions in different cell types and at different stages of disease remains limited. Further investigations on the genetic and molecular mechanisms of CD40 involved in SCZ development will be crucial for reconciling the disparities observed across different study methodologies and shedding light on the nuanced interactions at play in SCZ pathogenesis.

In the forward MR analysis of this study, three key instrumental variables were identified, including rs117191103, rs12624433, and rs7250371. Among these, rs12624433 is located in the intronic region of the SLC12A5 gene, which encodes the KCC2 protein. KCC2 plays a critical role in maintaining chloride homeostasis and GABAergic signaling in the central nervous system, processes essential for balancing neuronal excitation and inhibition (E/I).⁶⁷ Disruption of KCC2 expression can disturb this balance, potentially reverting neurons to an immature physiological state.⁶⁸ Such a state may adversely affect neuronal development, synaptic connectivity, and normal neural function,^68,69 thereby increasing the genetic risk for psychiatric disorders. Previous studies have also reported altered KCC2 expression in brain samples from patients with psychiatric disorders, such as decreased KCC2 protein levels in the dorsal lateral prefrontal cortex⁷⁰ and hippocampus⁷¹ of SCZ patients. Additionally, SLC12A5 is located in chromosome 20q13.12, a region that has been linked to both SCZ⁷² and BD.⁷³ Based on these findings, we hypothesize that the KCC2 protein, encoded by the SLC12A5 gene, may be involved in the genetic and neurodevelopmental pathology of SCZ and BD. And rs12624433 might influence the synthesis and expression of KCC2 protein by regulating SLC12A5 gene expression or affecting mRNA splicing, thereby providing protective effects against SCZ and BD. While these findings underscore the functional significance of rs12624433, further research is necessary to validate these hypotheses and explore their broader implications in cognitive and immune traits.

In reverse MR analysis, we observed an association between a genetically determined elevated risk of TS and decreased levels of Flt3L. Currently, there is a lack of clinical research exploring the relationship between Flt3L and TS. However, in the largest GWAS of TS published to date, it has been confirmed that the Fms-related tyrosine kinase 3 (Flt3), the receptor for Flt3L, exhibits the strongest association with TS.⁷⁴ Further investigations are needed in the future to reveal the effect of TS on Flt3L.

Another noteworthy finding in our investigation is that the protective effect of CD40 on SCZ was partially attributed to its regulation of FA values in the superior frontal occipital fasciculus. Reduced FA values in the left SFOF have been reported in patients with SCZ.⁷⁵ Another study indicated that SCZ patients who experience auditory verbal hallucination exhibited reduced FA value in the right SFOF.⁷⁶ Furthermore, a negative association has been observed between FA values in the SFOF of both cerebral hemispheres and the susceptibility to SCZ,²¹ which is further supported by our study. Nonetheless, our finding ought to be interpreted with caution because only the IVW method supported a statistically positive correlation between CD40 and the FA value in the bilateral SFOF, while the remaining four MR analysis methods did not support the conclusion. In addition, as a result of the insufficiently detailed clinical information available in existing GWAS datasets, whether this mediating effect is only for SCZ patients with auditory verbal hallucinations or if it is applicable to all SCZ patients is currently unclear. Further investigation is required in this field moving forward.

It is worth noting that in order to obtain robust results, we employed conservative Bonferroni corrections for multiple testing and rigorously removed instrumental variables that were strongly associated with potential confounders, in which case some previously reported causality may not be preserved in our study. For instance, several observational studies^77,78 and meta-analyses^79,80 support the association of increased IL-6 with SCZ, and there is an MR analysis⁸¹ that further points to a causal relationship between the two; our study did not find this correlation. This is attributed to our application of a rigorous P-value cutoff for selecting genetic instruments, which led to the retention of only one SNP (rs2228145) in IL-6. This SNP, which was significantly associated with rheumatoid arthritis among autoimmune diseases, was therefore removed in the step of removing confounders. We found the same immune-protective effect of CD40 on SCZ and BD, whereas no such relationship was found in other psychiatric disorders, probably because of the similar genetic and biological profiles of SCZ and BD.^82,83 Our study provides new evidence for the similarity of immune abnormalities between SCZ and BD, further supporting the viewpoint that closely associated disorders exhibit comparable patterns of inflammatory alterations.⁷⁹

Admittedly, this study has several limitations. First, to mitigate population stratification bias, all GWAS participants were exclusively of European descent, which may limit the applicability of our findings to populations with different ethnic backgrounds. Second, the lack of detailed phenotype information in the available GWAS data prevented us from exploring associations with specific case subtypes, such as first-episode and treatment-resistant cases. Third, we could not assess the timing relevance of inflammatory proteins on psychiatric outcomes due to the varying ages at which the exposures were measured in adults. Additionally, the measurement of inflammatory markers in the GWAS was taken at a single time point, providing only static snapshots. Consequently, these levels could be confounded by recent infections, the time of day they were measured, and other transient factors. This limitation should be acknowledged, as it may impact the interpretation of the association between inflammatory proteins and psychiatric outcomes. Furthermore, while we made considerable efforts to include five known confounders in our analysis, proving the independence assumption in MR studies is inherently complex. Therefore, the potential influence of unknown confounders cannot be completely ruled out.

In conclusion, we used the latest GWAS data to systematically assess bidirectional causality between 91 inflammatory proteins and 11 psychiatric disorders. We found CD40 exerts an immunoprotective effect on SCZ and BD, and the effect of CD40 on SCZ was partially mediated by modulating FA values in the superior frontal occipital fasciculus. In the future, it is crucial to replicate and expand upon our findings by utilizing larger-scale and more detailed GWAS data. Additionally, further experimental analyses are essential for delving deeper into the causality, potential roles, and therapeutic potential of inflammatory proteins in brain diseases.

Supplementary material

Supplementary material is available at https://academic.oup.com/schizophreniabulletin/.

Author contributions

ZCL and LLZ designed the study. LLZ completed the data collection and performed the data analyses. LLZ, LWT, WQL, SJZ, AJL prepared all figures and tables. LLZ completed the manuscript writing. LY, YH and XGC revised the manuscript. ZCL supervised this study. All authors contributed to this study.

Funding

This study is supported by the National Research and Development Grant from the Ministry of Science and Technology (Grant No. 2021YFE0191400 (X.C.)), National Natural Science Foundation of China (Grant No. 82101576 (Z.L.)), Science and Technology Innovation Program of Hunan Province (Grant No. 2022RC1040 (Z.L.)), Natural Science Foundation of Hunan Province of China (Grant No.2022JJ40695(Z.L.)), and the Fundamental Research Funds for the Central Universities of Central South University (Grant No. 2024ZZTS0932 (S.Z.)).

Data availability

The custom codes employed in this study can be accessed on GitHub at https://github.com/zhaolinlinnn/MR-analysis. All software packages used are publicly available.

Disclosures

The authors declare no potential conflicts of interest or financial interests related to this study.